Unfreeze operators as soon as possible due to peers' expirations #581
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
nolar
changed the title
|
This pull request introduces 1 alert when merging 3c9b757 into 5563bdb - view on LGTM.com new alerts:
|
nolar
changed the title
nolar
force-pushed
the
peering-repr-jitter
branch
from
e2d18fc to
914a3db
Compare
This should distribute the peering keepalives evenly after some time of functioning. There is already such an implicit jitter — the `PATCH` operation with varying duration (due to its networking nature). But it is not controllable, and the distribution in time happens slowly. With multiple namespaces served (each with its own peering keepalive), it would be better to distribute them faster to reduce the peak load on the K8s API.
Previously, the unfreezing only happened on the next occurrence when the peering resource was updated. Since most blocking peers were gone by that time, the update only happened from the operator's own keepalive change. Which lead to operators not unfreezing for up to 1 minute by default. With this change, the operators will estimate how long it is left till the blocking peers are supposed to expire, and —if there were no other signals or activities from other peers— will unfreeze immediately. The unfreeze is triggered by a fake update on the peering resource for the same operator — i.e. the same as keepalive would do in he next 0..60 seconds. But this is not a problem in case the peering was completely idle for some time. This also ensures that the unfreeze will be re-evaluated by the actual state of the peering resource, not an in-memory assumption about it.
nolar
force-pushed
the
peering-repr-jitter
branch
from
914a3db to
38e2f8d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Unfreeze operators when the blocking peers (other operators of the same "peering neighbourhood") are supposed to expire, not when the next keepalive happens as it was before.
This will reduce the waiting time for up to 60 seconds to much lower delay. However, delays of 60 seconds are still possible but less probable: if the blocking operators have sent their keepalives moments before exiting, they will be believed to exist for the next 60 second.
Besides, add a little jitter to the peering keepalives, so that they would distribute evenly after some time of functioning. There is already such an implicit jitter — the
PATCHoperation with varying duration (due to its networking nature). But it is not controllable, and the distribution in time happens slowly. With multiple namespaces served (each with its own peering keepalive), it would be better to distribute them faster to reduce the peak load on the K8s API.And also some code cleanups: improve reprs of peers for readability; remove the constructor types that are not needed anymore (peers are used only internally, and we do not feed
timedelta/datetimeinto them).This is an extraction of needed, useful, but unrelated changes from a bigger PR.